xref: /qemu/block/linux-aio.c (revision 7a4e543d) 
1 /*
2  * Linux native AIO support.
3  *
4  * Copyright (C) 2009 IBM, Corp.
5  * Copyright (C) 2009 Red Hat, Inc.
6  *
7  * This work is licensed under the terms of the GNU GPL, version 2 or later.
8  * See the COPYING file in the top-level directory.
9  */
10 #include "qemu/osdep.h"
11 #include "qemu-common.h"
12 #include "block/aio.h"
13 #include "qemu/queue.h"
14 #include "block/raw-aio.h"
15 #include "qemu/event_notifier.h"
16 
17 #include <libaio.h>
18 
19 /*
20  * Queue size (per-device).
21  *
22  * XXX: eventually we need to communicate this to the guest and/or make it
23  *      tunable by the guest.  If we get more outstanding requests at a time
24  *      than this we will get EAGAIN from io_submit which is communicated to
25  *      the guest as an I/O error.
26  */
27 #define MAX_EVENTS 128
28 
29 #define MAX_QUEUED_IO  128
30 
31 struct qemu_laiocb {
32     BlockAIOCB common;
33     struct qemu_laio_state *ctx;
34     struct iocb iocb;
35     ssize_t ret;
36     size_t nbytes;
37     QEMUIOVector *qiov;
38     bool is_read;
39     QSIMPLEQ_ENTRY(qemu_laiocb) next;
40 };
41 
42 typedef struct {
43     int plugged;
44     unsigned int n;
45     bool blocked;
46     QSIMPLEQ_HEAD(, qemu_laiocb) pending;
47 } LaioQueue;
48 
49 struct qemu_laio_state {
50     io_context_t ctx;
51     EventNotifier e;
52 
53     /* io queue for submit at batch */
54     LaioQueue io_q;
55 
56     /* I/O completion processing */
57     QEMUBH *completion_bh;
58     struct io_event events[MAX_EVENTS];
59     int event_idx;
60     int event_max;
61 };
62 
63 static void ioq_submit(struct qemu_laio_state *s);
64 
65 static inline ssize_t io_event_ret(struct io_event *ev)
66 {
67     return (ssize_t)(((uint64_t)ev->res2 << 32) | ev->res);
68 }
69 
70 /*
71  * Completes an AIO request (calls the callback and frees the ACB).
72  */
73 static void qemu_laio_process_completion(struct qemu_laio_state *s,
74     struct qemu_laiocb *laiocb)
75 {
76     int ret;
77 
78     ret = laiocb->ret;
79     if (ret != -ECANCELED) {
80         if (ret == laiocb->nbytes) {
81             ret = 0;
82         } else if (ret >= 0) {
83             /* Short reads mean EOF, pad with zeros. */
84             if (laiocb->is_read) {
85                 qemu_iovec_memset(laiocb->qiov, ret, 0,
86                     laiocb->qiov->size - ret);
87             } else {
88                 ret = -EINVAL;
89             }
90         }
91     }
92     laiocb->common.cb(laiocb->common.opaque, ret);
93 
94     qemu_aio_unref(laiocb);
95 }
96 
97 /* The completion BH fetches completed I/O requests and invokes their
98  * callbacks.
99  *
100  * The function is somewhat tricky because it supports nested event loops, for
101  * example when a request callback invokes aio_poll().  In order to do this,
102  * the completion events array and index are kept in qemu_laio_state.  The BH
103  * reschedules itself as long as there are completions pending so it will
104  * either be called again in a nested event loop or will be called after all
105  * events have been completed.  When there are no events left to complete, the
106  * BH returns without rescheduling.
107  */
108 static void qemu_laio_completion_bh(void *opaque)
109 {
110     struct qemu_laio_state *s = opaque;
111 
112     /* Fetch more completion events when empty */
113     if (s->event_idx == s->event_max) {
114         do {
115             struct timespec ts = { 0 };
116             s->event_max = io_getevents(s->ctx, MAX_EVENTS, MAX_EVENTS,
117                                         s->events, &ts);
118         } while (s->event_max == -EINTR);
119 
120         s->event_idx = 0;
121         if (s->event_max <= 0) {
122             s->event_max = 0;
123             return; /* no more events */
124         }
125     }
126 
127     /* Reschedule so nested event loops see currently pending completions */
128     qemu_bh_schedule(s->completion_bh);
129 
130     /* Process completion events */
131     while (s->event_idx < s->event_max) {
132         struct iocb *iocb = s->events[s->event_idx].obj;
133         struct qemu_laiocb *laiocb =
134                 container_of(iocb, struct qemu_laiocb, iocb);
135 
136         laiocb->ret = io_event_ret(&s->events[s->event_idx]);
137         s->event_idx++;
138 
139         qemu_laio_process_completion(s, laiocb);
140     }
141 
142     if (!s->io_q.plugged && !QSIMPLEQ_EMPTY(&s->io_q.pending)) {
143         ioq_submit(s);
144     }
145 }
146 
147 static void qemu_laio_completion_cb(EventNotifier *e)
148 {
149     struct qemu_laio_state *s = container_of(e, struct qemu_laio_state, e);
150 
151     if (event_notifier_test_and_clear(&s->e)) {
152         qemu_bh_schedule(s->completion_bh);
153     }
154 }
155 
156 static void laio_cancel(BlockAIOCB *blockacb)
157 {
158     struct qemu_laiocb *laiocb = (struct qemu_laiocb *)blockacb;
159     struct io_event event;
160     int ret;
161 
162     if (laiocb->ret != -EINPROGRESS) {
163         return;
164     }
165     ret = io_cancel(laiocb->ctx->ctx, &laiocb->iocb, &event);
166     laiocb->ret = -ECANCELED;
167     if (ret != 0) {
168         /* iocb is not cancelled, cb will be called by the event loop later */
169         return;
170     }
171 
172     laiocb->common.cb(laiocb->common.opaque, laiocb->ret);
173 }
174 
175 static const AIOCBInfo laio_aiocb_info = {
176     .aiocb_size         = sizeof(struct qemu_laiocb),
177     .cancel_async       = laio_cancel,
178 };
179 
180 static void ioq_init(LaioQueue *io_q)
181 {
182     QSIMPLEQ_INIT(&io_q->pending);
183     io_q->plugged = 0;
184     io_q->n = 0;
185     io_q->blocked = false;
186 }
187 
188 static void ioq_submit(struct qemu_laio_state *s)
189 {
190     int ret, len;
191     struct qemu_laiocb *aiocb;
192     struct iocb *iocbs[MAX_QUEUED_IO];
193     QSIMPLEQ_HEAD(, qemu_laiocb) completed;
194 
195     do {
196         len = 0;
197         QSIMPLEQ_FOREACH(aiocb, &s->io_q.pending, next) {
198             iocbs[len++] = &aiocb->iocb;
199             if (len == MAX_QUEUED_IO) {
200                 break;
201             }
202         }
203 
204         ret = io_submit(s->ctx, len, iocbs);
205         if (ret == -EAGAIN) {
206             break;
207         }
208         if (ret < 0) {
209             abort();
210         }
211 
212         s->io_q.n -= ret;
213         aiocb = container_of(iocbs[ret - 1], struct qemu_laiocb, iocb);
214         QSIMPLEQ_SPLIT_AFTER(&s->io_q.pending, aiocb, next, &completed);
215     } while (ret == len && !QSIMPLEQ_EMPTY(&s->io_q.pending));
216     s->io_q.blocked = (s->io_q.n > 0);
217 }
218 
219 void laio_io_plug(BlockDriverState *bs, void *aio_ctx)
220 {
221     struct qemu_laio_state *s = aio_ctx;
222 
223     s->io_q.plugged++;
224 }
225 
226 void laio_io_unplug(BlockDriverState *bs, void *aio_ctx, bool unplug)
227 {
228     struct qemu_laio_state *s = aio_ctx;
229 
230     assert(s->io_q.plugged > 0 || !unplug);
231 
232     if (unplug && --s->io_q.plugged > 0) {
233         return;
234     }
235 
236     if (!s->io_q.blocked && !QSIMPLEQ_EMPTY(&s->io_q.pending)) {
237         ioq_submit(s);
238     }
239 }
240 
241 BlockAIOCB *laio_submit(BlockDriverState *bs, void *aio_ctx, int fd,
242         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
243         BlockCompletionFunc *cb, void *opaque, int type)
244 {
245     struct qemu_laio_state *s = aio_ctx;
246     struct qemu_laiocb *laiocb;
247     struct iocb *iocbs;
248     off_t offset = sector_num * 512;
249 
250     laiocb = qemu_aio_get(&laio_aiocb_info, bs, cb, opaque);
251     laiocb->nbytes = nb_sectors * 512;
252     laiocb->ctx = s;
253     laiocb->ret = -EINPROGRESS;
254     laiocb->is_read = (type == QEMU_AIO_READ);
255     laiocb->qiov = qiov;
256 
257     iocbs = &laiocb->iocb;
258 
259     switch (type) {
260     case QEMU_AIO_WRITE:
261         io_prep_pwritev(iocbs, fd, qiov->iov, qiov->niov, offset);
262 	break;
263     case QEMU_AIO_READ:
264         io_prep_preadv(iocbs, fd, qiov->iov, qiov->niov, offset);
265 	break;
266     /* Currently Linux kernel does not support other operations */
267     default:
268         fprintf(stderr, "%s: invalid AIO request type 0x%x.\n",
269                         __func__, type);
270         goto out_free_aiocb;
271     }
272     io_set_eventfd(&laiocb->iocb, event_notifier_get_fd(&s->e));
273 
274     QSIMPLEQ_INSERT_TAIL(&s->io_q.pending, laiocb, next);
275     s->io_q.n++;
276     if (!s->io_q.blocked &&
277         (!s->io_q.plugged || s->io_q.n >= MAX_QUEUED_IO)) {
278         ioq_submit(s);
279     }
280     return &laiocb->common;
281 
282 out_free_aiocb:
283     qemu_aio_unref(laiocb);
284     return NULL;
285 }
286 
287 void laio_detach_aio_context(void *s_, AioContext *old_context)
288 {
289     struct qemu_laio_state *s = s_;
290 
291     aio_set_event_notifier(old_context, &s->e, false, NULL);
292     qemu_bh_delete(s->completion_bh);
293 }
294 
295 void laio_attach_aio_context(void *s_, AioContext *new_context)
296 {
297     struct qemu_laio_state *s = s_;
298 
299     s->completion_bh = aio_bh_new(new_context, qemu_laio_completion_bh, s);
300     aio_set_event_notifier(new_context, &s->e, false,
301                            qemu_laio_completion_cb);
302 }
303 
304 void *laio_init(void)
305 {
306     struct qemu_laio_state *s;
307 
308     s = g_malloc0(sizeof(*s));
309     if (event_notifier_init(&s->e, false) < 0) {
310         goto out_free_state;
311     }
312 
313     if (io_setup(MAX_EVENTS, &s->ctx) != 0) {
314         goto out_close_efd;
315     }
316 
317     ioq_init(&s->io_q);
318 
319     return s;
320 
321 out_close_efd:
322     event_notifier_cleanup(&s->e);
323 out_free_state:
324     g_free(s);
325     return NULL;
326 }
327 
328 void laio_cleanup(void *s_)
329 {
330     struct qemu_laio_state *s = s_;
331 
332     event_notifier_cleanup(&s->e);
333 
334     if (io_destroy(s->ctx) != 0) {
335         fprintf(stderr, "%s: destroy AIO context %p failed\n",
336                         __func__, &s->ctx);
337     }
338     g_free(s);
339 }
340